10 1109@access 2019 2903166

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2903166, IEEE Access
DRAFT 1
Hybrid Precoding for Massive mmWave MIMO

Systems
Xianru Liu, Xueming Li, Shu Cao, Qingyong Deng, Rong Ran, Kien Nguyen and Tingrui Pei
Abstract—Due to high costs and power consumptions, fully To alleviate the shortage of frequency resource, millimeter
digital baseband precoding schemes are usually prohibitive in wave (mmWave) is deemed as a potential radio frequency
millimeter-wave massive MIMO systems. Therefore, hybrid pre- candidate. Due to its gigabit-per-second data rates [6]–[9],
coding strategies become the promising solutions. In this paper,
we present a novel real-time yet high performance precoding mmWave will be deployed in the future fifth generation (5G)
strategy. Specifically, the eigenvectors corresponding to the larger wireless communication systems. Compared to the currently
eigenvalues of the right unitary matrix after singular value used frequency bands, the main characteristic of mmWave
decomposition (SVD) on an array response matrix are used to communication is its tremendous increase in carrier frequency.
abstract the angle information of an analog precoding matrix. It implies that mmWave signals experience an orders-of-
As the obtained eigenvectors correspond to the larger singular
values, the major phase information of channels is captured. In magnitude fading while propagating through the space. There-
this way, the iterative search process for obtaining the analog fore, it is difficult to ensure enough signal-to-noise ratio
precoding vectors is avoided, and thus hybrid precoding can be (SNR) for the conventional MIMO transceiver architecture.
realized in parallel. To further improve its spectral-efficiency, we Fortunately, the substantially reduced wavelength enables the
enlarge the resultant vector set by involving more relevant vectors facility to put large antenna arrays in a much smaller space.
in term of their correlation values with the unconstrained optimal
precoder, and a hybrid precoder is thus produced by using the Hence, mmWave systems can integrate massive multiple-
vector set. Simulation results show that our proposed scheme input multiple-output (MIMO) transceiver elements to enhance
achieves near the same performance as the orthogonal matching signal gain and spectral efficiency. Consequently, massive
pursuit (OMP) does, whereas it costs much less complexities than mmWave MIMO architecture has been adopted as a standard
the OMP, and thus can be realized in parallel. in 5G communication systems [10]–[12], and becomes one of
Index Terms—Millimeter wave communication, MIMO, wire- the research hotspots in the wireless communication domain.
less communication, hybrid precoding For a massive mmWave MIMO system, to decrease the
complexity of the receiver and cancel the interference among
I. I NTRODUCTION data streams, signals should be precoded before transmission.
When fully digital baseband precoding is used, it is not
S the capacity demand on wireless communication is
A increasing rapidly, the communication systems operating
in microwave band are difficult to meet the quality-of-service
difficult to precode multiple data streams to remove the inter-
signal interference. However, an individual radio frequency
(RF) chain is required for each antenna, which is impractical
(QoS) requirement. Hence, in the past several years, many when the number of antennas is large. On the other hand, when
advanced technologies such as relay cooperation, diversity a pure analog precoding mode is taken, RF chain can be shared
multiplexing and cognitive radio are utilized to enhance spec- among different antennas, and the cost in terms of hardware
tral efficiency [1]–[5]. However, compared to the explosive and power consumption can be reduced greatly. However, it is
increasing demand on system capacity, the obtained gain is too only applicable to the scenarios with single data stream, so it
limited. Accordingly, exploring other radio frequency bands is not beneficial to improve spectral efficiency. Consequently,
seems to be the essential solution to mitigate the crowded the methodology of combining the digital and analog modes
frequency bands. to form a hybrid precoding scheme is introduced into massive
X. Liu is with the School of Information Science and Engineering, Central mmWave MIMO systems. It not only determines precoders’
South University, Changsha, 410086, China, e-mail: liuxianru@csu.edu.cn. structure, but also affects system performances [13]. Hence,
X. Li is with School of Information and Communication Engineering, a hybrid precoder which can achieve good performance has
Beijing University of Posts and Telecommunications, Beijing, 100876, China,
e-mail: lixm@bupt.edu.cn. attracted a lot of research attentions.
S. Cao and T. Pei are with the Key Laboratory of Hunan Province for In the past several years, much research are carried out
Internet of Things and Information Security, Xiangtan University, Xiangtan, around massive MIMO. In [14], the minimal requested number
411105, China, e-mail: 1192001987@qq.com; peitingrui@xtu.edu.cn.
Q. Deng and is with School of Information and Communication Engineer- of RF chains and phase shifters in massive MIMO system is
ing, Beijing University of Posts and Telecommunications, Beijing, 100876, analyzed. For distributed massive MIMO systems, the benefits
China, and the Key Laboratory of Hunan Province for Internet of Things and of beamforming training scheme are analyzed in [15], and
Information Security, Xiangtan University, Xiangtan, 411105, China, e-mail:
dengqingyong@xtu.edu.cn. its spectral efficiency with linear beamforming under pilot
R. Ran is with the School of Electrical and Computer Engineering, Ajou contamination is presented in [16]. For mixed-ADC/DAC
University, Suwon 16499, South Korea, e-mail: sunnyran@ajou.ac.kr. multipair massive MIMO relaying systems, both exact and
Kien Nguyen is with Graduate School of Engineering, Chiba University,
Chiba 263-8522, Japan, e-mail: nguyen@chiba-u.jp. approximate closed-form expressions of the achievable rate
Corresponding author: Q. Deng (dengqingyong@xtu.edu.cn) are derived and a power scaling law is presented in [17].
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2903166, IEEE Access
DRAFT 2
For the similar system model, when two-way relaying with each antenna array one by one, a near-optimal iterative hybrid
hardware impairments is considered, an approximation of the precoding scheme is proposed in [31] that is based on a more
spectral efficiency with maximum ratio processing is derived realistic subarray structure. An orthogonal matching and local
in closed form in [18]. In [19], the design of directional hybrid search method is proposed to implement analog precoding in
precoding for multiuser mmwave communication systems with [32]. Based on the structure of mmWave channels, the design
multiple eavesdroppers is investigated. To maximize the sum of a hybrid precoder is formulated as a sparsity constrained
rate and the energy efficiency of massive MIMO systems, a least squares problem, and a precoding algorithm is proposed
codebook based hybrid precoding scheme with low complexity to approximate the optimal unconstrained precoder in [33]. A
is proposed in [20]. hybrid precoding processor based on parallel-index-selection
In mmWave massive MIMO systems, the essence of hybrid and orthogonal matching pursuit without matrix inversion is
precoding is to use two constrained matrices to approach proposed in [34]. For multiuser massive MIMO systems, a
the fully digital precoding matrix or inverted channel matrix. hybrid block diagonalization scheme is proposed to approach
Accordingly, the intuitive solution of hybrid precoding is to the capacity performance of the traditional baseband digital
decompose the fully digital precoding/channel matrix under method in [35]. By exploring the idea of turbo equalizer
given constraints. By directly decomposing the predesigned together with the tabu search algorithm, in [36], a novel
unconstrained digital precoder/combiner, a hybrid RF and hybrid precoding scheme is proposed to achieve near-optimal
baseband precoding/combining scheme for multistream trans- performance. In [37], after the relevance of directional precod-
mission is proposed in [21]. In [22], an iterative precoding al- ing structures is established as a low-complexity and robust
gorithm is proposed for mmWave systems, where all precoding solution to meet the demands of data rate, a simple class
operations are done in analog domain, and mmWave system of directional schedulers is proposed based on the channel
can adapt the rank of its transmission in response to varying directional structure, and then the performance comparison
propagation conditions. In [23], a hybrid connected structure between the single-user and multi-user scenarios is provided.
for hybrid beamforming is proposed, where precoding matrix Though the performance of the aforementioned hybrid pre-
optimization problem is decomposed into multiple precoding coding schemes is very close to the optimal unconstrained
sub-matrix optimization problems and near-optimal hybrid precoding, most of them require an iterative search. Hence,
digital and analog precoders are designed through factorizing the real-time performance is not very good. In this paper, after
the precoding sub-matrix for each sub-array. In [24], channel analyzing the relationship between antenna response matrix
matrix is decomposed into an angle domain basis matrix and its right singular-value-decomposition (SVD) matrix, the
and a gain matrix, which perfectly matches the structure most related vectors are selected out, which serves as analog
of hybrid precoding. In [25], it is shown that the Gram precoder after each element is normalized. To further improve
matrix of frequency-selective channel can be decomposed into system spectral efficiency, the selected vector set is enlarged
frequency-flat and frequency-selective components, which can such that the most relevant base vector can be involved in
be used for the analog and baseband precoders, respectively. term of the correlation with the fully digital precoding matrix.
All these research work are focused on decomposing the In this way, the iterative search manner is not required any
channel matrix of the mmWave large-scale MIMO systems longer, and then the hybrid precoding scheme can achieve
through some mathematical methods. As the constraints of high spectral efficiency with a parallel manner. The main
analog matrix should be considered during decomposition, the contributions of this paper are summarized as follows.
computational complexity is very high.
To decrease the complexity of hybrid precoder design, • As a baseline of hybrid precoder design, SVD is adopted
some low-complexity hybrid precoding schemes have been to obtain the phase information of mmWave large-scale
proposed. A new view of multi-user hybrid massive mMIMO MIMO channel. After SVD is carried out on the array
systems from array signal processing perspective is presented response matrix, the right resultant matrix consists of
in [26], where a hybrid precoding and channel estimation a group orthogonal basis, which can be used to form
strategies are proposed. In [27], an explicit analytical sum-rate array response vectors. Hence, the problem of designing
expression for generic channel covariance-based beamform- hybrid precoding is converted into searching the most
ing scheme is derived, and a low-complexity joint statistical relevant basis vectors, which is reflected on the value of
beamforming and user scheduling algorithm is proposed. In eigenvalues, and can be realized easily.
[28], exploiting the spatial structure of mmWave channels, • Instead of taking advantage of the traditional optimization
precoding/combining is formulated as a sparse reconstruction techniques used in hybrid precoding, such as OMP,
problem, and a precoding algorithm is developed to accurately OMP+LS, the proposed hybrid precoding scheme utilizes
approximate optimal unconstrained precoder and combiner SVD on array response matrix, and the most relevant
with low-cost RF hardware. Depending on the known ar- antenna response vectors are selected as analog precoding
ray geometry and a low training and feedback, a hybrid matrix directly after being normalized. In this way, the
analog/digital precoding algorithm is proposed in [29]. An iterative search process, which is a major ingredient that
angle-domain hybrid precoding is proposed in [30], and a leads high computational complexity, is avoided, and a
novel channel tracking strategy is also presented. After the hybrid precoder can be obtained directly. Consequently,
capacity optimization problem is decomposed into a series hybrid precoding is promsing to realize in parallel mode.
of subproblems that are easier to be handled by considering • Unlike the schemes in [31], [32], [28], where the analog
DRAFT 3
precoding vectors are obtained one by one through an achieved by an analog phase shifter only adjusts the phases
rather than the amplitudes of signals, (FRF FRF )l,l = Nt−1
(i) (i)∗
iterative OMP, the proposed scheme can select out all the
analog precoding vectors simultaneously. Consequently, (i)
should hold for any i, where FRF denotes the ith column
the execution time should be short so that the massive vector of FRF and (.)l,l denotes the lth diagonal element of
mmWave MIMO system becomes more real-time. a matrix. Meanwhile, the transmitted signals should guarantee
• The accuracy of hybrid precoding relative to the OMP the power constraint, which results in ||FRF FBB ||2F = Ns .
scheme can be controlled through setting the number For narrowband block-fading propagation scenarios, the
of radio frequency chains. As we know, if all the right received signals at the receiver can be shown as (1), where
normalized eigenvectors corresponding to nonzero eigen- y is the NM S × 1 received signal vector, H is the channel
values are selected out as analog precoding vectors, all matrix with NM S ×NBS dimensions, ξ is the average received
the antenna response vectors can be reconstructed. Hence, power, and n is the additive white Gaussian noise (AWGN)
the required number of radio frequency chains can be vector following i.i.d distribution CN(0, σn2 ).
determined according to the requested accuracy. √
The rest of the paper is organized as follows. In Section II, y = ξHFRF FBB s + n. (1)
after the system model of mmWave large-scale MIMO is Without loss of generality, we assume that the channel
briefly introduced, the hybrid precoding problem is formu- matrix H is known at both the transmitter and receiver 1 .
lated. The proposed hybrid precoding scheme in parallel and At the receiver(s), the received signals are processed in the
its improved version are presented specifically in Section III. similar way to recover the original transmitted data streams.
We evaluate the performance of our solution schemes in Sec- Accordingly, the finally processed signals can be shown as
tion IV. Finally, Section V concludes the paper and presents (2), where WBB and WRF denote the baseband and analog
our future research. combiners, respectively.
Notations: A is a matrix; a is a vector; a or A is a
′
scalar; A(i) or A(i) is the ith column of A; (•) , (•)∗ , and √ ∗ ∗ ∗
(•)−1 denote the transpose, conjugate transpose, and inversion, ŷ = ξWBB WRF HFRF FBB s + WBB WRF ∗ n. (2)
respectively; E[•] denotes expectation operation.
Similar to the precoder, the analog combiner WRF at
the receiver only provides the phase adjustments. So the
II. S YSTEM D ESCRIPTION
elements of WRF should have the same amplitudes, i.e.
(WRF WRF )l,l = Nr−1 . Given the system model (2), its
In this section, we first describe our considered system (i) (i)∗
model. Then, the characteristic of mmWave MIMO channel spectral efficiency is described as (3) [28], where Rn =
is analyzed. ∗ ∗
σn2 WBB WRF WRF WBB is the noise covariance matrix after
the received signals are combined at the receiver. Therefore,
A. System Model for mmWave massive MIMO systems, the task of hybrid
We consider a single-user mmWave system in which one precoding is to maximize R through designing FBB , FRF ,
base station with NBS antennas transmits Ns data streams WBB and WRF jointly.
to a receiver with NM S antennas simultaneously. To support
the function of transferring multiple data streams in a parallel ξ −1 ∗ ∗
R = log2 (|INs + R WBB WRF HFRF FBB
mode, the transmitter and receiver are equipped with multiple Ns n
radio frequency (RF) chains. The number of RF chains are × F∗BB F∗RF H∗ WRF WBB |).
denoted as NRF and MRF , and they satisfy Ns ≤ NRF ≤ (3)
NBS and Ns ≤ MRF ≤ NM S constraints, respectively.
In this paper, we focus on the precoder design at the
B. Cluster-based Channel Model
transmitter due to the reciprocities between the transmitting
process and receiving process. The similar approach can be Due to the characteristics of severe path-loss and high level
utilized in obtaining the combiner at the receiver, and is of correlation among antennas, the traditional channel model
omitted for brevity. For the Ns transmitted data streams, before is not applicable for mmWave large-scale MIMO transceiver
going through RF chains, they should be precoded in the array. Consequently, a narrow-band clustered channel repre-
baseband. Denote the baseband precoder, analog precoder and sentation based on the extended Saleh-Valenzuela model is
transmitted symbol vector as FBB , FRF , and s, respectively, usually adopted [39]–[41], where the channel matrix H is
and their dimensions are NRF × Ns , NBS × NRF and assumed to be a sum of the contributions of Ncl different
Ns × 1. Suppose the transmit power of s is normalized, i.e., clusters, each of which has Nray propagation paths. Accord-
E[ss∗ ] = N1s INs . After precoding in the baseband domain, ingly, the channel model of mmWave massive MIMO systems
the resultant vector given as FBB s is the input of NRF RF can be abstracted as (4), where γ is a normalization factor,
chains for the up conversion. Next, an analog precoder FRF αil is the complex gain of the lth propagation path in the
is applied for adjusting phase/angle to maximize the system ith cluster. Functions Λr (•) and Λt (•) represent the gains of
capacity or minimize the interference according to the system 1 For practical mmWave massive MIMO systems, the channel state informa-
requirements. Therefore, the discrete-time transmitted signals tion (CSI) can be obtained through channel estimation technology at receiver,
is finally represented as x = FRF FBB s. As the precoder FRF and feedback to transmitter.
DRAFT 4
transmit and receive antenna, respectively, that correspond to Fopt . The commonly used method of designing FRF is to
different departure and arrival angles. ar (•) and at (•) denote select the most relevant array response vectors. As a result,
the normalized respective receive and transmit array response we formulate the optimization problem as [28],
vectors when azimuth and elevation angles vary, and they are
independent [28]. Moreover, ϕril (θil r
) and ϕtil (θil t
) present the arg min {||Fopt − At F̂BB ||F }
At ,F̂BB
arrival and departure azimuth (elevation) angles, respectively, (7)
of the lth ray in the ith cluster. subject to ||diag(F̂BB F̂∗BB )||0 = NRF
∑ ||At F̂BB ||2F = Ns .
t ∗
H=γ αi,l Λr (ϕril , θil
t
)Λt (ϕril , θil
t
)ar (ϕril , θil
t
)at (ϕril , θil ) .
i,l where At is the antenna response matrix with the dimension of
(4) NBS × Ncl Nray (Ncl and Nray are the numbers of scattering
Note that when uniform planar array structure is considered, cluster and propagation paths per cluster, respectively, in
both ar (•) and at (•) can be expressed as the format given as clustered channel model, which will be described briefly in
(5), where W and H are the numbers of antenna elements, the next subsection), and ||•||0 denotes pseudo norm returning
k = 2π λ (λ is the corresponding wavelength) and d is the the number of non-zero elements. Implicitly, by searching the
inter-element spacing between two neighbor antennas, N = most suitable antenna response vectors in At , the constraints
W H is the size of the antenna array. For simplicity, we use of the analog precoder can be neglected.
At and Ar to represent the matrices consisting of all array Once the analog precoder FRF FBB has been designed,
response vectors at (•) and ar (•) respectively. These matrix the combiner WBB WRF at the receiver can be obtained via
representations will be used in the sequel. the similar approach. Let us assume Wopt = WBB WRF
1 and relax the hardware limitations, the optimal Wopt can
aU P A (ϕ, θ) = √ [1, · · · , ekd(msin(ϕ)sin(θ)+ncos(θ)) , be achieved by minimizing the mean squared error (MMSE)
N
· · · , ekd((W −1)sin(ϕ)sin(θ)+(H−1)cos(θ)) ]T . between the transmitted and processed received signals [38].
(5) Then, the problem of designing hybrid combiners can be
formulated as (8), where y is the received signal vector, and
III. H YBRID P RECODING S TRATEGY ΩRF is a constant-gain phase-only entries with NM S × MRF
size.
In this section, after analyzing the relation between the
channel matrix and antenna array response vectors, the prob- arg min {||E[yy∗ ]1/2 Wopt − WRF WBB ||F }
WRF ,WBB
lem of hybrid precoding is formulated. Then, a novel low-
subject to WRF ∈ ΩRF .
complexity hybrid precoding strategy and its improved version
(8)
are proposed.
Note that the process to solve (8) and (6) at the transmitter
A. Problem Formulation and receiver are similar, and hence we only present the strategy
To maximize R in (3), FBB , FRF , WBB and WRF should to design hybrid precoder in this paper, and the method to
be optimal. Nevertheless, due to the non-convex constraints optimize the hybrid combiners is omitted for the sake of
of FRF and WRF , it is extremely difficult to tackle the brevity.
optimization problem maximizing R via FRF , WBB and Note that the column vectors of the fully digital pre-
WRF . Consequently, the problem of designing analog/digital coding matrix Fopt are orthogonal [28]. However, they are
precoders and combiners is usually decoupled into two in- not necessarily complete orthogonal bases for the channel’s
dependent sub-problems. One is designing precoders for the row space since some eigenvectors corresponding to non-zero
transmitter, another is designing combiners for the receiver, eigenvalues are not selected out due to the constraint in the
and both of them follow the similar approach to obtain number column vectors (only NS column vectors are taken
solutions. At the transmitter, for data streams s, the achievable out). On the other hand, from (4), it can be seen that the
mutual information of the mmWave massive MIMO system linear combination of all array response vectors at (ϕtil , θilt
)
can be shown as (6). forms the channel matrix H. Hence, it can be concluded that
the linear combination of at (ϕtil , θil t
) can be used to present
ξ ∗ ∗ ∗
I(FRF , FBB ) = log2 (|I + HF F F
RF BB BB RF F H |). F opt accurately. Therefore, we can argue that hybrid precoder
NS σn2 design is to find two smaller matrices FRF and FBB such that
(6)
Fopt = FRF FBB . Moreover, FBB should be unit-amplitude
To maximize I(FRF , FBB ) in (6), the classical methods
which implies that the column vectors of FBB is able to
usually consist of two stages. First, considering the entirety of
present Fopt . Design (7) can be equally converted into the
FRF FBB and omitting all hardware limitations, the mutual
following problem
information can be maximized by Fopt = V(:, 1 : Ns ), where
∗
the SVD of H is UΣV . Fopt is the fully optimal baseband
precoding matrix, which consists of the Ns column vectors (Fopt opt
RF , FBB ) =arg min||Fopt − FRF FBB ||F ,
in V that correspond to the highest singular values in Σ. (i)
subject to FRF ∈ {at (ϕtil , θil t
), ∀i, l}, (9)
Second, considering the hardware constraints, FRF and FBB
are designed using some optimization strategies to approach ||FRF FBB ||F = Ns .
2
DRAFT 5
B. Solution Strategy RF chains is limited 2 , the construction of At is probably not

perfect. Consequently, its performance is degraded compared
To solve the optimization problem in (9), its essence is to
to the optimal iterative search scheme.
select out the NRF most relevant antenna response vectors
Based on the above analysis, the proposed hybrid precoding
from {at (ϕtil , θil
t
), ∀i, l}, which is denoted as At for notational
scheme can be summarized as Algorithm 13
simplicity. In the existing hybrid precoding schemes, optimal
matching strategies are commonly used, such as [28] [32].
Algorithm 1 Low-complexity hybrid precoding algorithm.
Though the performance of these hybrid precoding schemes
1: Obtain Fopt through SVD on H.
is very close to that of the fully digital baseband precoding
2: A′t = UΣV∗
scheme, its computational complexity is exponentially increas-
3: FRF = [V1 , V2 , · · · , VNRF ]∗ .
ing with the number of transceiver antennas, which leads to
4: FRF = norm(FRF )
huge latency. In fact, for the antenna response matrix At , if an
5: FBB = (F∗ RF FRF )
−1 ∗
FRF Fopt
orthogonal basis can be used to present it, the corresponding
6: return FRF , FBB
phase information is obtained, and an approximation analog
precoder can be constructed. In this way, hybrid precoding
can be realized without using any iterative search process, and For mmWave massive MIMO transceiver, as NRF << NBS
computational complexity can be reduced greatly. and NRF << Ncl Nray , it is incomplete that the NRF most
For antenna response matrix A′t , when it is decomposed relevant orthogonal basis vectors are used to approximate
using SVD, it can be shown as A′t = UΣV∗ , where the antenna response matrix At . Furthermore, in the OMP
the dimensions of A′t , U, Σ and V are Ncl Nray × NBS , hybrid precoding schemes, just the NRF relevant antenna
Ncl Nray × Ncl Nray , Ncl Nray × NBS and NBS × NBS , response vectors in At are selected out as the hybrid precoder.
respectively. To see more clear relationship between A′t and Consequently, there is a gap between the performance of
V, let us assume β = UΣ. Then, let β i , Ai and Vi denote Algorithm 1 and that of the fully digital baseband precoding
the ith row/row/column of β, A′t and V, respectively, for the scheme. As a compromise, an enlarged candidate orthogonal
sake of convenience. Consequently, the following relation can basis set is considered next, which means more orthogonal
be obtained. basis vectors are selected according to the correlation with the
fully digital baseband precoder Fopt . This idea is detailedly
∗
Ai = [β i V1∗ β i V2∗ , · · · , β i VN BS
∗
] = β i [V1∗ V2∗ , · · · , VN BS
]. demonstrated in Algorithm 2, where n is the parameter used
(10) to set the scale of the candidate orthogonal basis vectors and
It implies that the orthogonal basis Vi∗ ∈ ∀i can be used NRF ≤ n ≤ NBS .
to express Ai . For (10), as β i can be removed into the
part of digital precoder, the remaining task of the hybrid Algorithm 2 The improved low-complexity hybrid precoding
precoding is to select out the most relevant orthogonal vectors process.
from V. Therefore, searching the most relevant vectors in 1: Require : Fopt .
At successively is equivalent to obtaining the most suitable 2: A′t = UΣV∗ .
orthogonal basis which can be realized in the parallel model. 3: F̃RF = [V1 , V2 , · · · , Vn ]∗ .
For Vi∗ , though its norm is one, the norms of its element is 4: Ψ = F̃′RF Fopt .
not necessary one. Hence, its elements should be normalized. 5: Ω = diag(ΨΨ∗ ).
let Vi∗ = [V1i , V2i , · · · , VNBS i ]. Then, it can be rewritten as 6: [Y, I] = sort(Ω,′ descend′ ).
7: F̂RF = F̃RF (:, I(1 : NRF )).
V1i V2i VNBS i 8: FRF = norm(FRF ˆ ).
[V1i , V2i , · · · , VNBS i ] =[ , ,··· ,
]×
|V1i | |V2i | |VNBS i | 9: FBB = (FRF FRF )−1 F∗
∗
RF Fopt .
 
|V1i | 0 ··· , 0 10: return FRF , FBB .
 0 |V 2i | ··· , 0 
 ,
 ··· ··· ··· , ··· 
0 0 · · · , |VNBS i |
(11) C. Real-time Analysis
To see the computational complexity of the proposed hybrid
VNBS i precoding scheme, we compare it with the classical hybrid
in which [ |V1i | , |V2i | , · · · , |VN i | ] can be realized with the
V1i V2i
 BS  precoding schemes in [28] and [32]. From Algorithm 1 and Al-
|V1i | 0 ··· , 0
 0 |V | · · · , 0  gorithm 2, it can be seen that most of the computing operations
analog phase shifter, and   ···
2i  are concentrated on the SVD, sorting and matrix inversion

··· ··· , ···
(but just be required once). The traditional hybrid precoding
0 0 · · · , |VNBS i |
can be removed from the part of baseband precoder. From the schemes, such as OMP and OMP+LS, usually consist of an
above analysis, if all the vectors corresponding to non-zero iterative search process, in which sorting, matrix inversion
eigenvalues are taken out for the construction of analog phase 2 It can be seen that the number of RF chains is equal to the number of the
shifter, the performance of the hybrid precoding in this way is selected vectors from At .
equal to that of the OMP scheme. Obviously, as the number of 3 norm(•) function means normalizing each elements of the matrix.
DRAFT 6
and normalization operations are repeated for each iteration, 20

and are time-consuming. Consequently, the proposed hybrid OMP [N s=1]
precoding is more real-time than the traditional schemes. OMP [N s=3]
Spectral Efficiency (bits/s/Hz)

Algorithm 1 [N s=1]
15
IV. P ERFORMANCE E VALUATION Algorithm 1 [N s=3]
Fully Digital Precoding [N s=1]
In this section, numerical simulations are presented to show
the performance of our proposed hybrid precoding schemes. In 10
Algorithm 2 [N s=1]
the simulation, the cluster-based channel model is considered,
Algorithm 2 [N s=3]
in which Ncl = 8 and there are Nray = 10 rays per cluster.
Both the arrival and departure azimuth and elevation angles 5
of each cluster follows the Laplacian distribution. Similar
to [28] and [42], the spread angle in azimuth and elevation
angles are also set as 20o and 60o , respectively. Moreover, 0
-40 -35 -30 -25 -20 -15 -10 -5 0
it is also assumed that all clusters have the equal transmit SNR (dB)
power. In practical mmWave massive MIMO systems, the base
stations are usually more capable to configure larger massive
Fig. 2. The spectral efficiency of four precoding schemes with the varying
transceiver antenna array compared to the mobile terminals. SNR, where Ns = 1 and Ns = 3 are set under the conditions NBS = 64,
Hence, in our simulation, we let the number of antennas at NM S = 16 and NRF = MRF = 8, respectively.
the base station be larger than that of the mobile user, and
the scenarios with NBS = 64, NM S = 16 and NBS = 256,
NM S = 64 are considered. It is not surprise that Algorithm 2 always outperforms Algo-
Fig. 1 and Fig. 2 show the spectral efficiency with the rithm 1. But we notice that their performance gap decreases
varying SNR under the condition that NBS = 64, NM S = 16 as the number of RF chain increases. The essential of our
for NRF = MRF = 4 and NRF = MRF = 8, respectively. proposed schemes is how to select the relevant basis vectors
For the purpose of comparison, the simulation results of the to construct a new matrix which can accurately represent the
fully digital precoding and OMP schemes are presented as antenna response matrix At . In Algorithm 1, the number of
well. In order to identify the impact of the number of data basis vectors of At , which is the same as that of RF chains,
streams on the system spectral efficiency, we also consider is used to produce an analogy precoder. Since the channel
Ns = 1 and 3 cases. matrix H consists of vectors of At , the optimal analogy
precoder should be heavily related to At . But the number of
20 RF chain limits the number of basis vector candidates which
OMP [N s=1]
we can select to construct the representative matrix, there is
OMP [N s=3]
not guarantee that the selected vectors can contribute enough
Algorithm 1 [N s=1]
15 to produce a good precoder. However, in Algorithm 2, a larger
Algorithm 1 [N =3]
s number of eigenvectors are allowed to be correlated with the
optimal precoder, so more vector candidates can be used to
Fully Digital Precoding [N =3]
10 s determine the most relevant basis vectors. Therefore, a better
Algorithm 2 [N =1]
s precoder is possible to be produced compared to Algorithm 1,
Algorithm 2 [N s=3]
which is confirmed by simulation results. Furthermore, when
5 the number of RF chains increases, the probability of selecting
the most relevant basis vectors becomes larger, and hence the
spectral efficiency gap is reduced.
0 To show the impact of antenna array size on the spectral
-40 -35 -30 -25 -20 -15 -10 -5 0
efficiency, the simulation results for the cases NBS = 256 and
SNR (dB)
NM S = 64 are presented in Fig. 3 and Fig. 4 for NRF = 4
and NRF = 8, respectively. In Fig. 3, we can observe that
Fig. 1. The spectral efficiency of four precoding schemes with the varying
SNR, where Ns = 1 and Ns = 3 are set under the conditions NBS = 64,
the performance gap between the proposed Algorithm 2 and
NM S = 16 and NRF = MRF = 4, respectively. the OMP gets bigger compared to Fig. 1 due to the increased
number of antennas. It is because that when the size of the
From Fig. 1 and Fig. 2, we can observe that the spectral antenna array increases, the number of transmission paths
efficiency achieved by Algorithm 1 and Algorithm 2 approach increase. Consequently, the components that construct At
to that of the OMP scheme with respect to the number of becomes more complex, and it is more difficult to construct At
RF chains. For example, in Fig. 2, with NRF = MRF = 8, by using the orthogonal vectors. However, when we increase
the spectral performance of Algorithm 2 is almost as same the number of RF chains, as illustrated in Fig. 4, Algorithm
as that of the OMP and Fully Digital Precoding schemes. 2 approaches the OMP for both Ns = 1 and 3 cases, while
That is because, the more RF chains are available, the more costs much less complexities.
basis vectors can be used to generate the more accurate In Fig. 5, we consider three cases that Ns = 1, 2 and 3
representative matrix of the antenna response matrix At . under the conditions that NBS = 64, NM S = 16 and NRF =
DRAFT 7
35
OMP [N s=1] 20
Algorithm 1 [N =1]
OMP [N s=3] s
30
Sprectral Efficiency (bits/s/Hz)

Algorithm 2 [N =1]
Algorithm 1 [N s=1] s
25 Algorithm 1 [N s=3] 15 Algorithm 1 [N =2]

s
Fully Digital Precoding [N s=1] Algorithm 2 [N s=2]

20 Fully Digital Precoding [N s=3] Algorithm 1 [N s=3]
Algorithm 2 [N s=1] 10 Algorithm 2 [N =3]
15 s
Algorithm 2 [N s=3]
10
5
5
0 0
-40 -35 -30 -25 -20 -15 -10 -5 0 -40 -35 -30 -25 -20 -15 -10 -5 0
SNR (dB) SNR (dB)
Fig. 3. The spectral efficiency of four precoding schemes with the varying Fig. 5. The spectral efficiency of the proposed hybrid precoding schemes
SNR, where Ns = 1 and Ns = 3 are set under the conditions NBS = 256, with the varying SNR, where Ns = 1, Ns = 2 and Ns = 3 are set under the
NM S = 64 and NRF = MRF = 4, respectively. conditions NBS = 64, NM S = 16 and NRF = MRF = 8, respectively.
35 30
OMP [N s=1] Algorithm 1 [N s=1]
OMP [N s=3] Algorithm 2 [N s=1]
30

25
Algorithm 1 [N s=1] Algorithm 1 [N s=2]
25 Algorithm 1 [N s=3] Algorithm 2[N s=2]
20
Fully Digital Precoding [N s=1] Algorithm 1 [N =3]
s
20 Fully Digital Precoding [N s=3] Algorithm 2 [N s=3]
15
Algorithm 2 [N s=1]
15
Algorithm 2 [N s=3]
10
10
5 5
0 0
-40 -35 -30 -25 -20 -15 -10 -5 0 -40 -35 -30 -25 -20 -15 -10 -5 0
SNR (dB) SNR (dB)
Fig. 4. The spectral efficiency of four precoding schemes with the varying Fig. 6. The spectral efficiency of the proposed hybrid precoding schemes
SNR, where Ns = 1 and Ns = 3 are set under the conditions NBS = 256, with the varying SNR, where Ns = 1, Ns = 2 and Ns = 3 are set under the
NM S = 64 and NRF = MRF = 8, respectively. conditions NBS = 256, NM S = 64 and NRF = MRF = 8, respectively.
MRF = 8. From simulation results, we can observe that the spectral efficiency near linearly increases as the number of
spectral efficiencies of the proposed two algorithms increase in data streams at the middle or high SNR regimes. Our proposed
term of the number of data streams, which follows the general Algorithm 2 behaves the same as the OMP when Ns is less
behaviors of MIMO systems. In Fig. 6, we preserve NRF = than 6, but starts to degrade when Ns is over 6, and the
MRF = 8 and only increase the numbers of transmit and proposed Algorithm 1 degrades from Ns = 2. It is not difficult
receive antennas up to 256 and 64, respectively. The spectral to understand that given At , the selected orthogonal vectors
efficiency of the proposed Algorithm 2 increases as the number which are relevant to Fopt , become the loose representatives
of data streams increases. But the proposed Algorithm 1 with of the vectors of Fopt when the number of data stream
Ns = 3 does not achieve the best performance. That is because increases. Therefore, the interference cancellation capability
the precoder via Algorithm 1 with the degree of freedom of of the precoder is degraded, which results in the performance
NRF = MRF = 8 is not robust enough to deal with additional loss. The issue can be addressed by increasing the number of
interference caused by the increased data streams. candidate vectors used to generate the precoder (i.e., Algorithm
Therefore, in order to learn the impact of the number of data 2), but the complexity is increased as well. There arises a
streams on the spectral efficiency thoroughly, we illustrate the tradeoff between the interference cancellation capability of the
spectral efficiencies with respect to the number of data streams precoder via Algorithm 2 and the number of basis vectors
when SNR is given as 0dB in Fig. 7. From simulation results, used to generate the representative matrix. Namely, a trade-
we can observe that the fully digital precoding scheme and off between performance and complexity. Fortunately, since
OMP follow the general behaviors of MIMO systems that the NRF ≥ 2Ns is required to approach the optimal according
DRAFT 8
to the results given in [14], our proposed Algorithm 2 is generally require an iterative and successive search processes
implementable for the most reasonable scenarios. to achieve the analog phase shifters, our proposed scheme
exploits the correlation values between the eigenvectors of
the array response matrix and the digital optimal precoder
60
in selecting the most relevant basis vectors which are used
to reconstruct the representative matrix of the array response
50
matrix. Therefore, the iterative search process is no longer
40 required and the proposed scheme can significantly reduce
the complexity and be realized in parallel. Meanwhile, the
30
OMP [SNR = 0dB]
proposed scheme achieves a comparable performance as OMP.
Algorithm 1 [SNR = 0dB]
20 Fully Digital Precoding [SNR = 0dB]
Algorithm 2 [SNR = 0dB] ACKNOWLEDGMENT
10 This work was supported by the Open Project Program
of the Key Laboratory of Universal Wireless Communica-
0 tions (2016-KFKT-2016104), Ministry of Education, the Bei-
1 2 3 4 5 6 7 8
The Number of Data Streams (Ns) jing University of Posts and Telecommunications, in part
by the Natural Science Foundation of China (61771414,
61711540306 and 61672447), and in part by the Natural
Fig. 7. The spectral efficiency of four precoding schemes with the increasing Science Foundation of Hunan Province of China (2017JJ2249
Ns , where SNR = 0dB is set under the conditions NBS = 256, NM S = 64
and NRF = MRF = 8, respectively. and 2018JJ1025).
Finally, we illustrate the impact of the number of candidate R EFERENCES

vectors used in producing precoders in Fig. 8, where we still
consider NBS = 256, NM S = 64, NRF = MRF = 8 [1] L. Mohjazi, M. Dianati, G. K. Karagiannidis, S. Muhaidat, M. A.-
Qutayre, ”RF-powered cognitive radio networks: technical challenges and
and Ns = 3. From simulation results, we can see that limitations,” IEEE Commun. Mag., vol. 53, no. 4, pp. 94-100, Apr. 2015.
when the number of candidate vectors is greater than 12, [2] S. Wang, R. Ruby, V. C. M. Leung, Z. Yao, ”A Low- Complexity
the spectral efficiencies achieved by Algorithm 2 approach Power Allocation Strategy to Minimize Sum-Source-Power for Multi-
User Single-AF-Relay Networks,” IEEE Trans. commun., vol. 64, no. 8,
to a constant. For NRF = MRF = 8, only four additional pp. 3275-3283, Aug. 2016.
vector candidates are added to the selection set and obtain [3] J. Mietzner, R. Schober, L. Lampe, W. H. Gerstacker, and P. A. Hoeher,
near-optimal performance, which causes negligible complexity ”Multiple-antenna techniques for wireless communicationsłA comprehen-
sive literature survey,” IEEE Commun. Surveys Tuts., vol. 11, no. 2, pp.
increments compared to Algorithm 1. 87-105, Second Quart., 2009.
[4] Q. Deng, Z. Li, J. Chen, F. Zeng, H.-M. Wang, L. Zhou, Y.-J. Choi,
”Dynamic Spectrum Sharing for Hybrid Access in OFDMA-Based Cog-
32 nitive Femtocell Networks,” IEEE Trans. Veh. Tech., vol. 67, no. 11, pp.
10830-10840, Nov. 2018.
30 [5] S. Wang, R. Ruby, V. C. M. Leung, Z. Yao, X. Liu, Z. Li, ”Sump-Power
Minimization Problem in Multisource Single-AF-Relay Networks: A New

28 Revisit to Study the Optimality,” IEEE Trans. Veh. Tech., vol. 66, no. 11,
pp. 9958-9971, Nov. 2017.
Algorithm 2 [SNR =-10dB]
26
[6] P. B. Papazian, G. A. Hufford, R. J. Achatz, and R. Hoffman, ”Study
Algorithm 2 [SNR =0dB]
of the local multipoint distribution service radio channel,” IEEE Trans.
Fully Digital Precoding [SNR =-10dB]
Fully Digital Precoding [SNR =0dB]
Broadcasting, vol. 43, no. 2, pp. 175-184, 1997.
24
OMP [SNR =-10dB] [7] T. S. Rappaport, G. R. MacCartney, Jr., M. K. Samimi, and S. Sun,
OMP [SNR =0dB] ”Wideband millimeter-wave propagation measurements and channel mod-
22 els for future wireless communication system design (invited paper),”
IEEE Trans. Commun., vol. 63, no. 9, pp. 3029C3056, Sep. 2015.
20 [8] R. Daniels and R. W. Heath, Jr., ”60 GHz wireless communications:
emerging requirements and design recommendations,” IEEE Veh. Technol.
18 Mag., vol. 2, no. 3, pp. 41-50, 2007.
8 9 10 11 12 13 14 15 16 [9] J. Andrews, T. Bai, M. Kulkarni, A. Alkhateeb, A. Gupta, and R. W.
The Number of Selected Array Response vectors used to Correlate Heath, Jr., ”Modeling and analyzing millimeter wave cellular systems,”
IEEE Trans. Commun., vol. 65, no. 1, pp. 403C430, Jan. 2017
[10] A. I. Sulyman, A. T. Nassar, M. K. Samimi, G. R. M. Jr., ”Radio
Fig. 8. The spectral efficiency with the increasing number of selected array propagation path loss models for 5G cellular networks in the 28 GHz
response vectors to correlate with the fully digital precoding matrix, where and 38 GHz millimeter-wave bands,” IEEE Commun. Mag., vol. 52, no.
SNR = −10dB and SNR = 0dB are set under the conditions NBS = 256, 9, pp. 78-86, Sep. 2014.
NM S = 64 and NRF = MRF = 8, respectively. [11] M. Elkashlan, T. Q. Duong, and H.-H. Chen, ”Millimeter-wave commu-
nications for 5G: Fundamentals: Part I [Guest Editorial],” IEEE Commun.
Mag., vol. 52, no. 9, pp. 52-54, 2014
[12] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C, K.
V. C ONCLUSION Soong, J. C. Zhang, What will 5G be? IEEE J. Sel. Areas Commun., vol.
In this paper, we propose a novel hybrid precoding scheme 32, no. 6, pp. 1065-1082, Jun. 2014
[13] J. Zhang, L. Dai, X. Li, Y. Liu, L. Hanzo. ”On Low-Resolution ADCs in
for a mmWave massive MIMO system with the cluster-based Practical 5G Millimeter-Wave Massive MIMO Systems,” IEEE Commun.
channel model. Unlike the existing precoding schemes, which Mag., vol. 56, no. 7, pp. 205-211, Jul. 2018.
DRAFT 9
[14] T. E. Bogale, L. B. Le, A. Haghighat, and L. Vandendorpe, ”On the [36] X. Gao, L. Dai, C. Yuen, and Z. Wang, ”Turbo-like beamforming based
number of RF chains and phase shifters, and scheduling design with on Tabu search algorithm for millimeter-wave massive MIMO systems,”
hybrid analog-digital beamforming,” IEEE Trans. Wireless Commun., vol. IEEE Trans. Veh. Tech., vol. 65, no. 7, pp. 5731-5737, Jul. 2016.
15, no. 5, pp. 3311-3326, May 2016. [37] V. Raghavan, S. Subramanian, J. Cezanne, A. Sampath, O. H. Koymen,
[15] J. Li, D. Wang, P. Zhu, X. You, ”Benefits of Beamforming Training and J. Li, ”Single-user versus multi-user precoding for millimeter wave
Scheme in Distributed Large-Scale MIMO Systems,” IEEE Access, vol. MIMO systems,” IEEE J. Sel. Areas Commun., vol. 35, no. 6, pp. 1387-
6, pp. 7432-7444, Mar. 2018. 1401, Jun. 2017.
[16] J. Li, D. Wang, P. Zhu, J. Wang, X. You, ”Downlink Spectral Efficiency [38] T. Kailath, A. H. Sayed, and B. Hassibi, Linear estimation. Prentice
of Distributed Massive MIMO Systems With Linear Beamforming Under Hall, 2000.
Pilot Contamination,” IEEE Trans. Veh. Tech., vol. 67, no. 2, pp. 1130- [39] H. Xu, V. Kukshya, and T. Rappaport, ”Spatial and temporal character-
1145, Feb. 2018. istics of 60-GHz indoor channels,” IEEE J. Sel. Areas Commun., vol. 20,
[17] J. Zhang, L. Dai, Z. He, B. Ai, O. A. Dobre. ”Mixed-ADC/DAC no. 3, pp. 620-630, Apr. 2002.
Multipair Massive MIMO Relaying Systems: Performance Analysis and [40] A. M. Sayeed, ”Deconstructing multiantenna fading channels,” IEEE
Power Optimization”. IEEE Trans. Commun., vol. 67, no. 1, pp. 140-153, Trans. Signal Process., vol. 50, no. 10, pp. 2563-2579, Oct. 2002.
Jan.2019. [41] V. Raghavan and A. M. Sayeed, ”Sublinear capacity scaling laws for
[18] J. Zhang, X. Xue, E. Bjrnson, B. Ai, S. Jin. ”Spectral Efficiency of Mul- sparse MIMO channels,” IEEE Trans. Inf. Theory, vol. 57, no. 1, pp.
tipair Massive MIMO Two-Way Relaying with Hardware Impairments.” 345-364, Jan. 2011.
IEEE Wireless Commun. Lett., vol. 7, no. 1, pp. 14-17, Feb. 2018. [42] Z. Pi and F. Khan, ”An introduction to millimeter-wave mobile broad-
[19] Y. Huang, J. Zhang, M. Xiao, ”Constant Envelope Hybrid Precoding band systems,” IEEE Commun. Mag., vol. 49, no. 6, pp. 101-107, Jun.
for Directional Millimeter-Wave Communications,” IEEE J. Sel. Areas 2011.
Commun., vol. 36, no. 4, pp. 845-859, Apr. 2018.
[20] S. He, J. Wang, Y. Huang, B. Ottersten, W. Hong, ”Codebook Based
Hybrid Precoding for Millimeter Wave Multiuser Systems,” IEEE Trans.
Signal Process., vol. 65 no. 20, pp. 5289-5304, Oct. 2017.
[21] W. Ni, X. Dong, and W.-S. Lu, ”Near-optimal hybrid processing for
massive MIMO systems via matrix decomposition,” IEEE Trans. Signal
Process., vol. 65, no. 15, pp. 3922-3933, Aug. 2017.
[22] O. E. Ayach, R. W. Heath, Jr., S. Rajagopal, Z. Pi, ”Multimode precoding
in millimeter wave MIMO transmitters with multiple antenaa sub-arrays,”
in Proc. IEEE Global Telecommun. Conf. (GLOBECOM), Atlanta, GA,
USA, Dec. 2013, pp. 3476-3480.
[23] D. Zhang, Y. Wang, X. Li, W. Xiang, ”Hybridly Connected Structure
for Hybrid Beamforming in mmWave Massive MIMO systems,” IEEE
Trans. Commun., vol. 66, no. 2, pp. 662-674, Feb. 2018.
[24] H. Lin, F. Gao, S. Jin, and G. Li, ”A new view of multi-user hybrid
massive MIMO: Angle division multiple access,” IEEE J. Sel. Areas
Commun., vol. 35, no. 10, pp. 2268-2280, Oct. 2017.
[25] W. M. Chan, T. Kim, H. Ghauch, M. Bengtsson, ”Subspace estimation
and hybrid precoding for wideband millimeter-wave MIMO systems,”
in Proc. 50th Asilomar Conference on Signals, Systems and Computers,
Pacific Grove, CA, USA, Nov. 2016, pp. 286-290.
[26] H. Lin, F. Gao, S. Jin, G. Y. Li, ”A New View of Multi-User Hybrid
Massive MIMO: Non-Orthogonal Angle Division Multiple Access,” IEEE
J. Sel. Areas Commun., vol. 35, no. 10, pp. 2268-2280, Oct. 2017.
[27] C. Zhang, Y. Huang, Y. Jing, S. Jin, L. Yang, ”Sum-Rate Analysis for
Massive MIMO Downlink with Joint Statistical Beamforming and User
Scheduling,” IEEE Trans. Wireless Commun., vol. 16, no. 4, pp. 2181-
2194, Apr. 2017.
[28] O. E. Ayach, S. Rajagopal, S. A.-Surra, , Z. Pi, R. W. H. Jr., ”Spatially
sparse precoding in millimeter wave MIMO systems,” IEEE Trans.
Wireless Commun., vol. 13, no. 3, pp. 1499-1513, Mar. 2014.
[29] A. Alkhateeb, R. W. Heath Jr., G. Leus, ”Achievable rates of multi-
user millimeter wave systems with hybrid precoding,” in Proc. IEEE
International Conf. Commun., London, UK, Jun. 2015, pp. 1232-1237.
[30] J. Zhao, F. Gao, W. Jia, S. Zhang, S. Jin, H. Lin, ”Angle Domain Hybrid
Precoding and Channel Tracking for mmWave Massive MIMO Systems,”
IEEE Trans. Wireless Commun., vol. 16, no. 10, pp. 6868-6880, Oct. 2017.
[31] L. Dai, X. Gao, J. Quan, S. Han, c.-L. I, ”Near-optimal hybrid analog
and digital precoding for downlink mmWave massive MIMO systems,”
in Proc. IEEE International Conf. Commun., London, UK, Jun. 2015, pp.
1334-1339.
[32] C. Rusu, R. M.-Rial, N. G.-Prelcic R. W. H. Jr., ”Low complexity hybrid
sparse precoding and combining in millimeter wave MIMO systems,” in
Proc. IEEE International Conf. Commun., London, UK, Jun. 2015, pp.
1340-1345.
[33] O. E. Ayach, R. W. H. Jr., S. A.-Surra, S. Rajagopal, Z. Pi, ”Low
complexity precoding for large millimeter wave MIMO systems,” in Proc.
IEEE International Conf. Commun., Ottawa, Canada, UK, Jun. 2012, pp.
3724-3729.
[34] Y.-Y. Lee, C.-H. Wang, and Y.-H. Huang, ”A hybrid RF/baseband
precoding processor based on parallel-index-selection matrix-inversion-
bypass simultaneous orthogonal matching pursuit for millimeter wave
MIMO systems,” IEEE Trans. Signal Process., vol. 63, no. 2, pp. 305-
317, Jan. 2015.
[35] W. Ni and X. Dong, ”Hybrid block diagonalization for massive multiuser
MIMO systems,” IEEE Trans. Commun., vol. 64, no. 1, pp. 201-211, Jan.
2016.

10 1109@access 2019 2903166

Uploaded by

Copyright:

Available Formats

You might also like

10 1109@access 2019 2903166

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

10 1109@access 2019 2903166

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in a future issue of this journal, but has not been

Hybrid Precoding for Massive mmWave MIMO

B. Solution Strategy RF chains is limited 2 , the construction of At is probably not

and normalization operations are repeated for each iteration, 20

precoding is more real-time than the traditional schemes. OMP [N s=3]

Spectral Efficiency (bits/s/Hz)

Sprectral Efficiency (bits/s/Hz)

25 Algorithm 1 [N s=3] 15 Algorithm 1 [N =2]

Fully Digital Precoding [N s=1] Algorithm 2 [N s=2]

Spectral Efficiency (bits/s/Hz)

Finally, we illustrate the impact of the number of candidate R EFERENCES

Minimization Problem in Multisource Single-AF-Relay Networks: A New

You might also like